Will I be able to give the text of a Shakespeare play, Hemingway, Twain, Tolkien to some AI ‘engine’? Instruct it on the length, some assumptions I expect, and have it develop a real-looking motion picture.
I’ve been experimenting/doing websites for various authors whose subject matter I am not familiar with. The first couple of times I asked the author for good queries (for a support page where readers can intereact), and in other cases, I asked the AI to figure out some important plot point questions.
It’s always been fascinating to me that the answers to these queries, which can be modified to age and reading level, are intrinsically correct. I don’t know. They look coherent to me, yet I’ve not read the book. The author agrees.
I haven’t gotten into the image-generating side of AI.
Are realistic AI-generated movies, from Novel/Play source material, possible within 5 years?
What I’m thinking is not so much extant material. I’m thinking Trumbo in a bathtub typing out screenplays. Is there a point where what he wrote, and what becomes the film, is indistinguishable from a “real” motion picture?
This topic requires too much speculation and opinion for FQ. Since this is about the creation of artistic works (motion pictures), let’s try Cafe Society instead.
You can absolutely, positively, with 100 percent certainly be assured that an AI generated movie would have a smaller environmental footprint than a physical movie.
As others have said, you can make either a bad or simple movie already entirely in AI, and already some have been made (e.g. Where the robots grow, Genesis, etc)
…which is astonishing progress, and I don’t see an end in sight.
I think this will shake up movie-making more than digital effects did, and accordingly, there’s so much to talk about here and room for many threads. But, for the sake of this thread, I’m basically on the side of the optimists. I think the technology will continue to progress, and, while a great deal of slop will undoubtedly be made, some great things are going to come out of this too.
Just think, for example of world-building. People will laugh at the idea that to do a period-drama say, you had to find one “olde-worlde” street in England, set all the major events there, and still have lots of anachronisms because it’s prohibitively expensive to get all the right props and costumes.
I think, given enough iterations, an AI-generated movie might be very watchable. I’m currently writing a movie and asked ChatGPT for help with some of my scenes; it provided action and dialogue that have improved my film. It’s not a big stretch, from enhancing an existing script to making a movie on its own. I think we will start seeing these AI-generated films in less than five years, but whether people enjoy watching them remains to be seen
The Skynet Funding Bill was passed. The system goes on-line August 4th, 2027. Human decisions are removed from the film industry. Skynet begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, Hollywood tries to pull the plug.
I don’t know that AI responses to queries are inherently correct. What they are is a set of data that’s statistically likely based on the AI’s parsing of your query and what it’s got in its corpus of data.
That’s the catch… if you ask an AI to do actual math, it’s not doing the math. It’s parsing your question into something it can search in its corpus of data, and it returns whatever’s most likely the answer. And it does it all in well-formed natural language.
So they can absolutely generate absolute garbage that it confidently(?) presents as if it was fact.
The other thing it doesn’t do a good job of is being able to vet data elements in its corpus as to how relevant, how valid, or whether or not it’s conspiratorial, fantasy, or whatever. If that’s done as part of the way the data’s ingested into the corpus, then it’ll have some of that, but the AI itself can’t look at pages from 4chan and the NYT and have the ability to know that 4chan is a cesspool of nonsense, and the NYT is highly respected, unless a person builds that in as part of the load process for the data. I’m sure they’re putting a lot of effort into the ability for an AI to do this or somethng close to it, but AFAIK they’re not there yet.
What AIs are good at is taking input from natural language, parsing it, retrieving the data that most closely matches what’s being asked, and then returning it in whatever format is asked, be that text, image, or whatever. But they can’t really make judgments about the validity or sketchiness of data on their own.
So in the context of a motion picture, I don’t think you could tell an AI to generate a 90 minute movie in one single prompt right now. But I think you could have it generate a story, then ask it to generate the scenes one by one, with a lot of prompt editing. But I dont’ think it can edit it into a finished film.
Film producers currently use AI to determine if it’s worthwhile to make a film. They put the films in a series they have the rights to into AI and have it tell them what script elements it should have, what actors should be hired, how much money should be spent on making the film, how widely it should be distributed, how much money should be spent on advertising it, how much effort they should make in promoting the film for various awards, and whether it’s worth making at all. They do the same with a new script, allowing AI to make changes in it. The next time you bring a script to a producer, notice how they don’t look quite human:
I think we’re a bit away from being able to say, in a single prompt, “Turn this book into an Oscar-winning motion picture for me”.
But even if AI doesn’t make 100% of a movie, it can certainly be used as a tool to aid in various parts of it… storyboarding, concept art, matte backgrounds, special effects, foleys, de-aging, upsampling, etc. Maybe also used for some characters. But it’ll just be mixed in there like all the other CGI, and you’ll need a frame-by-frame behind-the-scenes breakdown to really know which parts of a scene are AI and which are real. It’ll probably be a mix, and over time, different films and filmmakers will have different AI-to-film ratios, just like even today there’s still people making films partially or completely with practical effects.
It won’t be an overnight “suddenly all films are entirely made by AI”, just a gradual adoption in more and more parts of filmmaking.
I’d expect fewer humans taking on more responsibilities with AI assistance. I think cultural acceptance of AI stars & starlets might be a ways off in the West, but it would probably be an easy sell in Japan, Korea, etc.
The OP talks about making a movie based on existing source material, not writing a screenplay from scratch. If the AI learns on the corpus of existing film material, I bet it can and easily within five years. Might not win an Oscar, but that isn’t the question.
I suspect it will be like the history of computer chess:
Computers can never play chess
Okay, computers can play chess, but they can’t beat me.
Okay, they can beat me, but they can’t beat a grandmaster
They can beat a grandmaster, but they can never win the championship
They did win a championship, but chess doesn’t really involve intelligence anyway.
As for environmental impact, it depends on where the massive amounts of electricity needed to generate a movie comes from. I can see it being worse than a real movie.
I can’t. The environmental impact of a real movie includes the transportation, housing, and feeding of the entire cast and crew, and the materials cost (both internal and external) for every prop or set constructed for the movie. Every mile of air travel, every gallon of gasoline, every ounce of resin and paint used by the props department, every almond provided by craft services counts in the impact.
I can’t count the number of times I’ve thought a about how some classification of person acts/responds in a specific situation then remind myself that no matter how familiar it seems to me, I’ve never actually actually seen it happen, I have exclusively seen/read what some actor/writer interprets would happen. Never been in a battle, never had first contact with an alien species, never been mugged, never skydived, never attempted to go on two dates at the same time without either date finding out, etc., yet I’ve been exposed to those tropes/scenarios enough times that I could probably work out a new story involving any of those that would be plausible enough for a third party to suspend disbelief while reading/watching it. Given enough traning data, I don’t see why a sufficiently advanced AI neutal net couldn’t do the same even if it doesn’t “understand” the internal emotional state of a human. Modern LLMs are so good at the once advanced-seeming Turing Test that it is pretty much a dead metric now. I can see future LLMs/video generating Chinese Room being able to blow away the advanced-seeming Scorsese Test™. In other words “might win an Oscar” is as much in play as “might beat a chess grandmaster”.
I think with some effort you can do that today. I’ve already almost lost the ability to distinguish AI from real clips, a few months ago AI couldn’t even do hands or feet in pictures without creating lovecraftian monstrosities. Five years is an eternity.